SPSS Practice: Assumptions and Normality
| Independent Variable [IV] (number of groups) | |||||
| Dependent Variable [DV] (measurement level) | Two Groups | Three + Groups | |||
| Independent (“unpaired”) | Dependent (“paired”) | Independent (“unpaired”) | Dependent (“paired”) | ||
| Categorical | Non-parametric Tests | Chi-square | McNemar’s | Chi-square | Cochran’s Q |
| Ordinal | Mann-Whitney U | Wilcoxon Signed ranks | Kruskal Wallis H | Friedman’s | |
| Interval / Ratio (continuous) | Parametric Tests | Independent t-test | Dependent t-test | ANOVA | RM-ANOVA |
“What is the effect of TREATMENT (IV) on our OUTCOME (DV) of interest?”
Example: TREATMENT independent groups (placebo versus drug), OUTCOME interval/ratio (blood pressure)
Example: TREATMENT dependent group (pre/post yoga therapy), OUTCOME ordinal (back pain levels)
Example: TREATMENT independent 3+ groups (yoga therapy, none, aerobics), OUTCOME categorical (pass/fail of driving test)
| Correlations |
| Phi coefficient |
| Spearman’s rho |
| Pearson’s r |
| Independent Variable (number of groups) | |||||
| Dependent Variable (measurement level) | Two Groups | Three + Groups | |||
| Independent (“unpaired”) | Dependent (“paired”) | Independent (“unpaired”) | Dependent (“paired”) | ||
| Categorical | Non-parametric Tests | Chi-square | McNemar’s | Chi-square | Cochran’s Q |
| Ordinal | Mann-Whitney U | Wilcoxon Signed ranks | Kruskal Wallis H | Friedman’s | |
| Interval / Ratio (continuous) | Parametric Tests | Independent t-test | Dependent t-test | ANOVA | RM-ANOVA |
STEP #1
Check what measurement level your DV is.
STEP #2
Choose the column related to the number Groups in your study.
STEP #3
Choose the column where intervention groups are either “paired” or “unpaired.”
STEP #4
Match your column with the row to find which test to run.
STEP #1
Look at your Dependent Variable or outcome.
The data that we are looking at here is from the instruments you used to measure the effect of your intervention. Maybe you chose to measure stress with a commonly used psychological questionnaire or maybe you measured cholesterol levels or test scores.
What is its measurement level?
Categorical (such as yes or no; dead or alive; pass or fail).
Ordinal (such as health status – poor, average, excellent).
Interval ratio (for instance blood pressure, cholesterol level, rates of infection, or workplace satisfaction scores on a scale of 0-100).
STEP #2
Next you will look for the column that corresponds to the number of groups you have for your Independent Variable (also called experimental or predictor variable).
Remember, the independent variable is the thing in your study that was controlled by you (such as a medical intervention, or training initiative, or implementation of a modified protocol) for the purpose of making a change on some outcome in the population you are studying.
So…how many groups were involved in this intervention?
For example, if you were testing the effect of an evidence-based training initiative on employee workplace satisfaction or happiness, you might be interested in comparing the training initiative in one group to no training in another group. Here, then, you would have two groups being studied.
But, maybe you wish to modify your intervention so that you also have training and massage compared to training alone, and both of these compared to no training or massage. Here you would have three groups where you would be measuring the DV (that is, workplace satisfaction).
STEP #3
Before we decide on the statistical test to use, we must examine another part of the Independent Variable columns that correspond to the number of groups you have in your study. We are now interested in a characteristic called Independent (unpaired) and Dependent (paired) groups.
To explain this, we will use the same evidence-based training initiative example from the previous slide. Let us assume you have two groups that you want to use in studying employee workplace satisfaction and happiness.
For the sake of this illustration, we will say that one group, which comes from a unit X at the hospital you selected, will be compared to unit Y that will not receive any training. Clearly, you now have two groups (one with training, and one without training) that you will be evaluating on your dependent variable called satisfaction and happiness. What is important here is to recognize that each group of individuals comes from entirely distinct units. While they might know each other, there is no particular way that they might influence each other on the satisfaction and happiness test. In fact, you have to be sure that you do not have a husband and wife split between these units because of their potential to influence each other. In this kind of study, you create two independent groups.
For a different study design, you may decide to use only one group, testing them for satisfaction and happiness before the training, and then right after the training. You would again have two groups being tested, but this time the groups are composed of the same people (that is, the same persons tested twice) and would therefore be considered “dependent.”
This step would appear to be the easiest. You have basically one test option that shows up at the intersection of the chosen IV column and DV rows. In reality, however, the selection of the appropriate inferential test is not quite so simple. But we will leave most of the exceptions to a statistician to figure out.
All dependent variables that are either ordinal or categorical in nature must have nonparametric testing. Interval or ratio data, on the other hand, may require a bit more evaluation since it can be tested with either parametric tests or nonparametric tests. Here is the most likely scenario you may encounter:
If you have determined that your data is at the interval or ratio type of measurement level (that is, continuous), you still must determine if it is distributed normally before you select the statistical test. In this course, you will only be responsible for testing to see if your interval or ratio data is normally distributed. Your readings and resources have more on this topic.
If your distribution test comes back saying that your outcome data is not normally distributed, then you will likely need to use a nonparametric test. To find the appropriate nonparametric equivalent tests for non-normal interval or ratio data, you simply move to the row just above the parametric test you selected. In other words, if you had selected an independent t-test for your intervalor ratio data, you would use the Mann-Whitney U test if the data turns out to fail the assumption of normal distribution.
STEP #4